Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method of routing data in a network comprising a plurality of switches (a-f) and a plurality of nodes ( 0 - 7 ), said plurality of nodes being distinct of said plurality of switches, said data coming from a first node and being destined for a second node, said method comprising: pre-recording in each of the plurality of switches of the network a routing table, the routing table comprising a list of destination switches which, as a different entity from one or more destination nodes, are directly interfaced with the one or more destination nodes, wherein each entry in the routing table does not include one or more routes to a destination node but one or more routes to a destination switch directly interfaced with the destination node; reception of the data by a first switch via an input port; determination of a destination of said data by the first switch; when the first switch is directly interfaced with the second node via a specific output port selection then routing of the data by said specific output port; otherwise identification of a switch, as the destination switch, to which the second node is directly interfaced, by comparison between the destination of the data and the list of pre-recorded destination switches; routing of the data to the identified switch via an output port of the first switch, in such a way that the data can be routed to said identified switch, the selection of the output port to the identified switch being performed by comparison between the identified switch and a routing table of the first switch, each entry of said routing table comprising: information enabling a switch interfaced with at least one node to be identified; a set of aggregated information describing at least one route by a specific port to said switch interfaced with the at least one node.
2. The method according to claim 1 , wherein for a plurality of possible routes to said switch interfaced with the at least one node, the selection of the output port is made as a function of the aggregated information, said information comprising a number of hops, congestion or weighting of links or nodes on a route enabling the switch interfaced with the at least one node to be reached via the specific port.
This invention relates to network routing optimization in packet-switched networks, specifically improving route selection in multi-path environments. The problem addressed is inefficient routing decisions that lead to suboptimal performance due to lack of real-time network state awareness, such as congestion, hop count, or link/node weighting. The method enhances routing by selecting output ports based on aggregated network information for multiple possible routes to a destination switch. This information includes the number of hops, congestion levels, and weighting factors of links or nodes along each route. By analyzing these metrics, the system determines the most efficient path to reach the destination switch through a specific port. This dynamic selection process ensures better load balancing, reduced latency, and improved overall network performance by avoiding congested or inefficient routes. The solution is particularly useful in complex networks where multiple paths exist between nodes, allowing switches to make informed decisions rather than relying on static or limited routing tables. The aggregated information provides a holistic view of network conditions, enabling adaptive routing that responds to real-time changes in traffic patterns or link availability. This approach optimizes resource utilization and minimizes packet loss or delay, enhancing the reliability and efficiency of data transmission.
3. The method according to claim 1 , wherein the network is a supercomputer topology.
A method for optimizing data transmission in a supercomputer network addresses the challenge of efficiently routing data across high-performance computing systems. Supercomputers often employ complex topologies, such as fat-tree, hypercube, or mesh networks, to handle massive data volumes and parallel processing demands. However, these topologies can suffer from congestion, latency, and inefficient resource utilization, particularly when managing dynamic workloads. The method involves dynamically adjusting data routing paths within the supercomputer network to minimize latency and maximize throughput. It analyzes real-time network traffic patterns, identifies bottlenecks, and reroutes data through alternative paths to balance load distribution. The method may also prioritize critical data packets to ensure timely delivery for time-sensitive applications. Additionally, it can adapt to changes in network topology, such as node failures or reconfigurations, by recalculating optimal paths on-the-fly. The method may integrate with existing network protocols or operate as a standalone optimization layer. It can be applied to various supercomputer architectures, including those with hierarchical or distributed memory systems. By dynamically optimizing routing, the method enhances overall system performance, reduces idle time, and improves energy efficiency in large-scale computing environments.
4. A computer program product executed on a memory medium, capable of being implemented within a computerized processing unit and comprising instructions for implementation of a method according to claim 1 .
This invention relates to a computer program product designed to optimize data processing in a computerized system. The product is stored on a memory medium and executed within a processing unit, enabling efficient data handling and computational tasks. The core functionality involves implementing a method that processes input data through a series of computational steps to generate an output. These steps include receiving input data, performing one or more processing operations on the data, and producing a result based on the operations. The processing operations may involve mathematical computations, data transformations, or other logical manipulations tailored to specific applications. The program product is structured to ensure compatibility with various hardware configurations and operating environments, allowing flexible deployment across different systems. By executing the method, the program enhances data processing efficiency, reduces computational overhead, and improves system performance. The invention addresses the need for reliable and adaptable software solutions that can handle diverse data processing requirements while maintaining accuracy and speed. The program product is particularly useful in applications requiring real-time data analysis, automated decision-making, or large-scale data management.
5. A switch in a network comprising a plurality of switches (a-f) and a plurality of nodes ( 0 - 7 ), said plurality of nodes being distinct of said plurality of switches, said switch comprising a plurality of input and output ports capable of receiving and routing data coming from a first node and destined for a second node; a pre-recorded routing table, the routing table comprising a list of destination switches which, as a different entity from one or more destination nodes, are directly interfaced with the one or more destination nodes, wherein each entry in the routing table does not include one or more routes to a destination node but one or more routes to a destination switch directly interfaced with the destination node; means capable of determining a destination of said data; means capable of selecting then routing the data by a specific output port, when the switch is directly interfaced with the second node via said specific output port; means capable of identifying a second switch, as the destination switch, to which the second node is directly interfaced, by comparison between the destination of the data and the pre-recorded list of destination switches; means capable of selecting then routing the data to the identified second switch via an output port, in such a way that the data can be routed to said identified second switch, the selection of the output port to the identified switch being achieved by means capable of comparing the identified second switch with entries from the pre-recorded routing table, each entry of said routing table comprising: information enabling the identification of a switch interfaced with at least one node; a set of aggregated information describing at least one route by a specific port to said switch interfaced with the at least one node.
This invention relates to network routing in a system comprising multiple switches and nodes, where nodes are distinct from switches. The problem addressed is efficient data routing in such networks, particularly when switches must forward data to destination nodes without direct connections. The system includes switches with input and output ports for receiving and routing data between nodes. Each switch contains a pre-recorded routing table that lists destination switches, not nodes, to which data should be forwarded. The routing table entries specify routes to switches directly interfaced with destination nodes, rather than the nodes themselves. When data arrives at a switch, the system determines its destination node and checks if the switch is directly connected to that node. If so, it routes the data directly. If not, the system identifies the destination switch interfaced with the target node by comparing the destination with the routing table. The data is then routed to the identified switch via an appropriate output port, selected by matching the destination switch with routing table entries. Each routing table entry includes switch identification and aggregated route information for specific ports leading to the switch. This approach simplifies routing by abstracting node-level details, improving efficiency in multi-switch networks.
6. The switch according to claim 5 , comprising means capable of selecting, for a plurality of possible routes to said switch interfaced with the at least one node, one output port based on the aggregated information, said information comprising a number of hops for the specific port, congestion or weighting of links or nodes on a route enabling the switch interfaced with the at least one node to be reached.
This invention relates to network routing in a switch interfaced with at least one node, addressing the problem of efficiently selecting optimal routes in a network with multiple possible paths. The switch includes means for selecting an output port from a plurality of possible routes based on aggregated information about the network. This information includes the number of hops required to reach the switch, as well as congestion levels or weighting factors associated with links or nodes along the route. By analyzing these factors, the switch can determine the most efficient or least congested path to forward data, improving network performance and reliability. The selection process considers dynamic conditions such as link congestion or node load, allowing the switch to adapt to changing network states. This approach enhances routing decisions by incorporating real-time network metrics, ensuring better resource utilization and minimizing latency. The invention is particularly useful in complex networks where multiple paths exist between nodes, requiring intelligent routing to optimize traffic flow.
7. A network of a supercomputer, comprising the plurality of switches (a-f) and the plurality of nodes ( 0 - 7 ), the plurality of switches being produced in accordance with claim 5 .
A supercomputer network architecture addresses the challenge of efficiently interconnecting multiple computing nodes to maximize data transfer rates and minimize latency. The network includes a plurality of switches and a plurality of nodes, where the switches are designed to facilitate high-speed, low-latency communication between the nodes. The switches are configured to route data packets dynamically, ensuring optimal path selection based on network traffic conditions. The nodes, which may be individual processors or memory units, are interconnected in a structured topology to enable parallel processing and distributed computing. The switches incorporate advanced routing algorithms to prevent congestion and ensure reliable data transmission across the network. This architecture is particularly suited for high-performance computing applications, such as scientific simulations, large-scale data analysis, and real-time processing tasks. The design emphasizes scalability, allowing additional nodes and switches to be integrated without compromising performance. By optimizing the interconnection between nodes, the network enhances computational efficiency and reduces bottlenecks, making it ideal for demanding computational workloads.
8. The method according to claim 1 , wherein each destination switch, included in the list of destination switches, is a leaf switch.
In the field of network communication, particularly in data center or cloud computing environments, efficient and scalable packet forwarding is critical for managing high volumes of traffic. A common challenge is ensuring that data packets are routed optimally through network switches, especially in multi-tier architectures where leaf switches serve as endpoints for end devices. Traditional methods may suffer from inefficiencies in identifying and managing destination switches, leading to suboptimal routing and increased latency. This invention addresses the problem by providing a method for packet forwarding in a network where each destination switch in a list of destination switches is a leaf switch. The method involves receiving a packet at an ingress switch, determining a list of destination switches for the packet, and forwarding the packet to the appropriate destination switches. The key innovation lies in the specific configuration where all destination switches in the list are leaf switches, ensuring that the packet is routed directly to the endpoints without unnecessary intermediate hops. This approach improves efficiency by reducing latency and minimizing network congestion, particularly in large-scale networks with numerous leaf switches. The method may also include additional steps such as encapsulating the packet with routing information or using a forwarding table to identify the correct destination switches. By restricting the destination switches to leaf switches, the method ensures that packets are delivered directly to their intended endpoints, enhancing overall network performance.
9. The method according to claim 1 , wherein the plurality of switches are separate devices than the plurality of nodes.
A system and method for managing network traffic involves a plurality of nodes interconnected by a plurality of switches, where the switches are separate physical devices from the nodes. The nodes may be computing devices, servers, or other network endpoints, while the switches handle routing and forwarding of data packets between these nodes. The method includes dynamically configuring the switches to optimize network performance, such as by adjusting routing paths, load balancing, or prioritizing certain types of traffic. The switches may also monitor network conditions, detect congestion, or identify faulty connections, and take corrective actions to maintain efficient data flow. The separation of switches from nodes allows for centralized control and scalability, enabling the network to adapt to changing demands without requiring modifications to the individual nodes. This approach improves reliability, reduces latency, and enhances overall network efficiency by leveraging dedicated switching hardware to manage traffic independently of the computing resources. The system may be applied in data centers, cloud computing environments, or other large-scale networks where dynamic traffic management is critical.
Unknown
April 14, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.